On the synthesis of specific image processing automata by a simulated annealing-based design space search

A novel approach to both automatic data path synthesis and architecture optimization of existing machines is reported. This approach features innovations in the concept of automatic synthesis as well as in the method used in the system implementation. This work exhibits two new features: (1) at the level of the automatic architecture synthesis process, the automatic design of specific image processing automata is done from emulation results of a more general-purpose machine for image processing applications; and (2) at the level of system implementation, taking into account the constraints of design (speed, area, etc.), the method used to search the design space is a stimulated-annealing-based algorithm. Therefore, systematic optimization of the architecture is performed, leading to the design of a 'child automaton' at the architectural level from two files: the hardware description of the emulator machine; and the control data-flow graph.<
>